Recover from nvidia installation problems

olinart · November 22, 2020, 9:50pm

My fedora 31 system running an NVIDIA GPU failed recently after partially running for several months. I was unable to find the problem source - it occurred when I rebooted and may have been triggered by a system update. Thanks for any help, I know this is messy.
My installation had used the rpmfusion howto and included cuda11.0. So I tried to uninstall nvidia and start over, again using the rpmfusion howto.

I find the GPU

bash-5.0# /sbin/lspci | grep -e VGA
00:02.0 VGA compatible controller: Intel Corporation UHD Graphics 620 (rev 07)
09:00.0 VGA compatible controller: NVIDIA Corporation GP107 [GeForce GTX 1050] (rev a1)
There is some evidence of the nvidia installation.
modinfo -F version nvidia
455.38
The kernel-devel modules are installed.
I disabled my epel repo since it seemed to have some conflicts with rpmfusion. My working repolist is
repo id repo name
adobe-linux-x86_64 Adobe Systems Incorporated
cernbox CERNBOX2.0 client repository (EOS backend)
copr:copr.fedorainfracloud.org:abn:throttled Copr repo for throttled owned by abn
cuda cuda
eos-fc-30 fc-30 EOS release from EOS project
fedora Fedora 31 - x86_64
fedora-modular Fedora Modular 31 - x86_64
google-chrome google-chrome
google-chrome-64 google-chrome - 64-bit
google-chrome-unstable google-chrome-unstable
rpmfusion-free RPM Fusion for Fedora 31 - Free
rpmfusion-free-updates RPM Fusion for Fedora 31 - Free - Updates
rpmfusion-nonfree-nvidia-driver RPM Fusion for Fedora 31 - Nonfree - NVIDIA Driver
skype-stable skype (stable)
updates Fedora 31 - x86_64 - Updates
updates-modular Fedora Modular 31 - x86_64 - Updates

I can no longer install the akmod for nvidia.

dnf install akmod-nvidia
Last metadata expiration check: 1:42:07 ago on Sun 22 Nov 2020 09:21:01 AM PST.
Error:
Problem: package nvidia-kmod-common-3:455.45.01-1.el8.noarch requires nvidia-driver = 3:455.45.01, but none of the providers can be installed

package akmod-nvidia-3:455.38-1.fc31.x86_64 requires nvidia-kmod-common >= 3:455.38, but none of the providers can be installed

package nvidia-driver-3:455.45.01-1.el8.x86_64 requires dnf-plugin-nvidia, but none of the providers can be installed

conflicting requests

package dnf-plugin-nvidia-1.1-1.el8.noarch is filtered out by modular filtering

nothing provides python(abi) = 3.6 needed by dnf-plugin-nvidia-1.1-1.el8.noarch

package dnf-plugin-nvidia-1.6-1.el8.noarch is filtered out by modular filtering

nothing provides python(abi) = 3.6 needed by dnf-plugin-nvidia-1.6-1.el8.noarch

package dnf-plugin-nvidia-1.9-1.el8.noarch is filtered out by modular filtering

nothing provides python(abi) = 3.6 needed by dnf-plugin-nvidia-1.9-1.el8.noarch

nothing provides python(abi) = 3.6 needed by dnf-plugin-nvidia-2.0-1.el8.noarch

package xorg-x11-drv-nvidia-3:455.38-1.fc31.x86_64 is filtered out by modular filtering

A possible clue for why I’ve lost contact with the GTX1050 gpu is a message in /var/log/messages ( which only runs on shutdown)

Nov 21 18:55:48 lenolin systemd[1]: Stopped Create Volatile Files and Directories.
Nov 21 18:55:49 lenolin dracut-initramfs-restore[2448]: /usr/sbin/setfiles: conflicting specifications for /run/initramfs/etc/modprobe.d/nvidia-installer-disable-nouveau.conf and /run/initramfs/usr/lib/modprobe.d/nvidia-installer-disable-nouveau.conf, using system_u:object_r:lib_t:s0.
Nov 21 18:55:49 lenolin systemd[1]: dracut-shutdown.service: Succeeded.
Nov 21 18:55:49 lenolin systemd[1]: Stopped Restore /run/initramfs on shutdown.
Anything to be gained by upgrading to fed33?

computersavvy · November 22, 2020, 11:04pm

First, you have, or at least had, the epel repo and the modular repos enabled so you have conflicting packages.

I would disable these repos

cernbox CERNBOX2.0 client repository (EOS backend)
eos-fc-30 fc-30 EOS release from EOS project
fedora-modular Fedora Modular 31 - x86_64
cuda cuda
google-chrome google-chrome
google-chrome-unstable google-chrome-unstable
updates-modular Fedora Modular 31 - x86_64 - Updates

then uninstall all nvidia related packages

 dnf erase '*nvidia*'

finally reinstall all the nvidia packages from rpmfusion

dnf install akmod-nvidia nvidia* xorg-x11-drv-nvidia*

This will give you all matching and no conflicting packages for nvidia, including cuda for the nvidia card.
When you next reboot things should be normal.

olinart · November 23, 2020, 3:16am

Thanks for the detailed reply computersavy. I carefully went through your procedure.
First problem:

bash-5.0# dnf erase ‘nvidia’
Error:
Problem: The operation would result in removing the following protected packages: kernel-core
(try to add ‘–skip-broken’ to skip uninstallable packages)
I’m worried that removing a protected kernel will brick my OS. Is this perhaps a kernel that gets recompiled for nvidia?
Proceeding anyway, second problem:
bash-5.0# dnf install akmod-nvidia nvidia* xorg-x11-drv-nvidia*
Last metadata expiration check: 0:00:43 ago on Sun 22 Nov 2020 06:59:25 PM PST.
Error:
Problem: package akmod-nvidia-3:455.38-1.fc31.x86_64 requires nvidia-kmod-common >= 3:455.38, but none of the providers can be installed

conflicting requests
package xorg-x11-drv-nvidia-3:455.38-1.fc31.x86_64 is filtered out by modular filtering
(try to add ‘–skip-broken’ to skip uninstallable packages)
Tried it anyway with skip-broken and

Running transaction
Preparing : 1/1
Installing : nvidia-texture-tools-2.0.8-21.fc31.x86_64 1/9
Installing : nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64 2/9
Installing : mesa-libGLES-19.2.8-2.fc31.x86_64 3/9
Installing : egl-wayland-1.1.5-3.fc31.x86_64 4/9
Installing : xorg-x11-drv-nvidia-libs-3:455.38-1.fc31.x86_64 5/9
Installing : nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64 6/9
Installing : nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64 7/9
Installing : xorg-x11-drv-nvidia-kmodsrc-3:455.38-1.fc31.x86_64 8/9
Installing : xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 9/9
Running scriptlet: xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 9/9
/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_adv_train.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_ops_train.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_cnn_train.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8 is not a symbolic link

/sbin/ldconfig: /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcudnn_adv_infer.so.8 is not a symbolic link
So the akmod failed. Wish me luck when I try to reboot!

olinart · November 23, 2020, 3:33am

Rebooting takes me to my original state of not accessing the nvidia gpu.

dnf info --installed ‘nvidia’
Installed Packages
Name : nvidia-query-resource-opengl
Version : 1.0.0
Release : 6.fc31
Architecture : x86_64
Size : 23 k
Source : nvidia-query-resource-opengl-1.0.0-6.fc31.src.rpm
Repository : @System
From repo : fedora
Summary : Querying OpenGL resource usage of applications using the NVIDIA
: OpenGL driver
URL : https://github.com/NVIDIA/nvidia-query-resource-opengl/
License : BSD
Description : A tool for querying OpenGL resource usage of applications using
: the NVIDIA OpenGL driver. Requires NVIDIA 387 or later.

Name : nvidia-query-resource-opengl-lib
Version : 1.0.0
Release : 6.fc31
Architecture : x86_64
Size : 18 k
Source : nvidia-query-resource-opengl-1.0.0-6.fc31.src.rpm
Repository : @System
From repo : fedora
Summary : Library for nvidia-query-resource-opengl
URL : https://github.com/NVIDIA/nvidia-query-resource-opengl/
License : BSD
Description : This package contains library for nvidia-query-resource-opengl.

Name : nvidia-texture-tools
Version : 2.0.8
Release : 21.fc31
Architecture : x86_64
Size : 574 k
Source : nvidia-texture-tools-2.0.8-21.fc31.src.rpm
Repository : @System
From repo : fedora
Summary : Collection of image processing and texture manipulation tools
URL : http://code.google.com/p/nvidia-texture-tools/
License : MIT
Description : The NVIDIA Texture Tools is a collection of image processing and
: texture manipulation tools, designed to be integrated in game
: tools and asset conditioning pipelines.
:
: The primary features of the library are mipmap and normal map
: generation, format conversion and DXT compression.
:
: DXT compression is based on Simon Brown’s squish library. The
: library also contains an alternative GPU-accelerated compressor
: that uses CUDA and is one order of magnitude faster.

Name : nvidia-texture-tools-devel
Version : 2.0.8
Release : 21.fc31
Architecture : x86_64
Size : 13 k
Source : nvidia-texture-tools-2.0.8-21.fc31.src.rpm
Repository : @System
From repo : fedora
Summary : Development libraries/headers for nvidia-texture-tools
URL : http://code.google.com/p/nvidia-texture-tools/
License : MIT
Description : Headers and libraries for development with nvidia-texture-tools.

Name : xorg-x11-drv-nvidia-cuda-libs
Epoch : 3
Version : 455.38
Release : 1.fc31
Architecture : x86_64
Size : 121 M
Source : xorg-x11-drv-nvidia-455.38-1.fc31.src.rpm
Repository : @System
From repo : rpmfusion-nonfree-nvidia-driver
Summary : CUDA libraries for xorg-x11-drv-nvidia
URL : http://www.nvidia.com/
License : Redistributable, no modification permitted
Description : This package provides the CUDA driver libraries.

Name : xorg-x11-drv-nvidia-kmodsrc
Epoch : 3
Version : 455.38
Release : 1.fc31
Architecture : x86_64
Size : 18 M
Source : xorg-x11-drv-nvidia-455.38-1.fc31.src.rpm
Repository : @System
From repo : rpmfusion-nonfree-nvidia-driver
Summary : xorg-x11-drv-nvidia kernel module source code
URL : http://www.nvidia.com/
License : Redistributable, no modification permitted
Description : Source tree used for building kernel module packages
: (xorg-x11-drv-nvidia-kmod) which is generated during the build of
: main package.

Name : xorg-x11-drv-nvidia-libs
Epoch : 3
Version : 455.38
Release : 1.fc31
Architecture : x86_64
Size : 248 M
Source : xorg-x11-drv-nvidia-455.38-1.fc31.src.rpm
Repository : @System
From repo : rpmfusion-nonfree-nvidia-driver
Summary : Libraries for xorg-x11-drv-nvidia
URL : http://www.nvidia.com/
License : Redistributable, no modification permitted
Description : This package provides the shared libraries for
: xorg-x11-drv-nvidia.

No xorg-x11-drv-nvidia which I believe is the required driver.
Any suggestions? It may be time to upgrade to fed33, but I’d like to fix this first.

computersavvy · November 23, 2020, 4:44pm

What you displayed is not the same as I suggested for the removal step, and it seems you still have repos that are unneeded and conflicting that are active - especially the modular repos.
No, there is no special kernel for F31 and nvidia unless you have gotten it from the epel repo where part of your existing nvidia stuff was installed from. Epel repo is not fedora and often causes conflicts. Most people recommend to not use epel with fedora because of frequent version and dependency mismatches.

First, disable all repos except

fedora
fedora-updates
rpmfusion-free
rpmfusion-free-updates
rpmfusion-nonfree
rpmfusion-nonfree-updates

Limiting the number of repos enabled will help avoid conflicts and the others can be re-enabled later if you wish
Then do the removal of all existing nvidia packages, avoiding dependencies. You also need to remove all the currently installed cuda packages since installing the nvidia drivers from rpmfusion will also install the related cuda drivers.

dnf erase '*nvidia*' --noautoremove

Note the “*” on either side of “nvidia” which does not appear in your reply above (probably because the code tag at the beginning is not on its own line)

If this comes up with the kernel-core package you noted above, then do

 dnf disto-sync

and reboot before trying again.

Then, still with only the six repos listed above, repeat the install using this command (modified from above).

dnf install akmod-nvidia nvidia* xorg-x11-drv-nvidia* kernel-devel kernel-headers

This should fix the conflicts. If not you can address each one individually. Removing each epel8 package individually is almost a necessity if these steps do not clear it out. I notice that in the install list you showed several packages that did not come from rpmfusion. Nothing should be installed from anywhere but the main fedora repo or rpmfusion.

When the install is complete you should only have these nvidia related packages installed.

dnf list installed '*nvidia*'
Installed Packages
akmod-nvidia.x86_64                                                          3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
nvidia-persistenced.x86_64                                                   3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
nvidia-settings.x86_64                                                       3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia.x86_64                                                   3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-cuda.x86_64                                              3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-cuda-libs.x86_64                                         3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-devel.x86_64                                             3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-kmodsrc.x86_64                                           3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-libs.x86_64                                              3:455.38-1.fc32                                     @rpmfusion-nonfree-updates

You also will have one kmod-nvidia package that the system will build for you once the install is done. That is the actual nvidia driver module, custom built for your kernel.

I personally permanently disable the modular repos since I have seen conflicts with modular vs the regular fedora repos and your errors show similar…

After getting the nvidia drivers working I would suggest a full system upgrade with “dnf upgrade --refresh” while still using only the six above repos so that you know everything is up to date fedora wise. That will be needed anyway should you do a system upgrade to F33 later. As I am sure you are aware, F31 goes past EOL this month.

olinart · November 23, 2020, 8:55pm

Thanks again, I appreciate all your experience. but still some problems.
I did disable the repos temporarily as you suggested. They were mostly apps for communicating with CERN.
I apologize for the overly detailed posting this time. Might be best to skip to the bottom and work back.

Current list is

dnf repolist
repo id repo name
fedora Fedora 31 - x86_64
rpmfusion-free RPM Fusion for Fedora 31 - Free
rpmfusion-free-updates RPM Fusion for Fedora 31 - Free - Updates
rpmfusion-nonfree-nvidia-driver RPM Fusion for Fedora 31 - Nonfree - NVIDIA Driver
updates Fedora 31 - x86_64 - Updates

I did the dnf distro-sync and removed all the epel packages that I found. They were mostly relics. Also removed the cuda-11.0 code in /usr/local that I had downloaded from nvidia as suggested in the rpmfusion howto. Perhaps this won’t be needed anymore?
I tried it first without the rpmfusion-nonfree-nvidia-driver repo, but the akmod was missing

dnf install akmod-nvidia nvidia* xorg-x11-drv-nvidia* kernel-devel kernel-headers
Last metadata expiration check: 1:56:43 ago on Mon 23 Nov 2020 10:02:35 AM PST.
No match for argument: akmod-nvidia
No match for argument: nvidia-bug-report.log.gz
No match for argument: xorg-x11-drv-nvidia*
Package kernel-devel-5.8.16-100.fc31.x86_64 is already installed.
Package kernel-devel-5.8.17-100.fc31.x86_64 is already installed.
Package kernel-devel-5.8.18-100.fc31.x86_64 is already installed.
Package kernel-headers-5.8.18-100.fc31.x86_64 is already installed.
Error: Unable to find a match: akmod-nvidia nvidia-bug-report.log.gz xorg-x11-drv-nvidia*

After adding the rpmfusion-nonfree-nvidia-driver repo

bash-5.0# dnf install akmod-nvidia
Last metadata expiration check: 0:11:23 ago on Mon 23 Nov 2020 12:03:26 PM PST.
Error:
Problem: package akmod-nvidia-3:455.38-1.fc31.x86_64 requires nvidia-kmod-common >= 3:455.38, but none of the providers can be installed

conflicting requests

package xorg-x11-drv-nvidia-3:455.38-1.fc31.x86_64 is filtered out by modular filtering
(try to add ‘–skip-broken’ to skip uninstallable packages)
bash-5.0# dnf install nvidia*
Last metadata expiration check: 0:11:51 ago on Mon 23 Nov 2020 12:03:26 PM PST.
Dependencies resolved.
===========================================================================================================================
Package Architecture Version Repository Size
===========================================================================================================================
Installing:
nvidia-query-resource-opengl x86_64 1.0.0-6.fc31 fedora 18 k
nvidia-texture-tools x86_64 2.0.8-21.fc31 fedora 168 k
nvidia-texture-tools-devel x86_64 2.0.8-21.fc31 fedora 13 k
Installing dependencies:
nvidia-query-resource-opengl-lib x86_64 1.0.0-6.fc31 fedora 14 k

Transaction Summary

Install 4 Packages

Total download size: 213 k
Installed size: 627 k
Is this ok [y/N]: y
Downloading Packages:
(1/4): nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64.rpm 37 kB/s | 18 kB 00:00
(2/4): nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64.rpm 28 kB/s | 14 kB 00:00
(3/4): nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64.rpm 53 kB/s | 13 kB 00:00
(4/4): nvidia-texture-tools-2.0.8-21.fc31.x86_64.rpm 156 kB/s | 168 kB 00:01

Total 139 kB/s | 213 kB 00:01
Running transaction check
Transaction check succeeded.
Running transaction test
Transaction test succeeded.
Running transaction
Preparing : 1/1
Installing : nvidia-texture-tools-2.0.8-21.fc31.x86_64 1/4
Installing : nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64 2/4
Installing : nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64 3/4
Installing : nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64 4/4
Running scriptlet: nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64 4/4
Verifying : nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64 1/4
Verifying : nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64 2/4
Verifying : nvidia-texture-tools-2.0.8-21.fc31.x86_64 3/4
Verifying : nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64 4/4

Installed:
nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64 nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64
nvidia-texture-tools-2.0.8-21.fc31.x86_64 nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64

Complete!
bash-5.0# dnf install --skip-broken akmod-nvidia xorg-x11-drv-nvidia*
Last metadata expiration check: 0:28:29 ago on Mon 23 Nov 2020 12:03:26 PM PST.
Dependencies resolved.

Problem: package akmod-nvidia-3:455.38-1.fc31.x86_64 requires nvidia-kmod-common >= 3:455.38, but none of the providers can be installed

conflicting requests

package xorg-x11-drv-nvidia-3:455.38-1.fc31.x86_64 is filtered out by modular filtering
==========================================================================================================================
Package Architecture Version Repository Size
==========================================================================================================================
Installing:
xorg-x11-drv-nvidia-cuda-libs x86_64 3:455.38-1.fc31 rpmfusion-nonfree-nvidia-driver 24 M
xorg-x11-drv-nvidia-kmodsrc x86_64 3:455.38-1.fc31 rpmfusion-nonfree-nvidia-driver 18 M
xorg-x11-drv-nvidia-libs x86_64 3:455.38-1.fc31 rpmfusion-nonfree-nvidia-driver 78 M
Skipping packages with broken dependencies:
akmod-nvidia x86_64 3:455.38-1.fc31 rpmfusion-nonfree-nvidia-driver 26 k

Transaction Summary

Install 3 Packages
Skip 1 Package

Total download size: 121 M
Installed size: 387 M
Is this ok [y/N]: y
Downloading Packages:
(1/3): xorg-x11-drv-nvidia-kmodsrc-455.38-1.fc31.x86_64.rpm 1.5 MB/s | 18 MB 00:12
(2/3): xorg-x11-drv-nvidia-cuda-libs-455.38-1.fc31.x86_64.rpm 1.0 MB/s | 24 MB 00:23
(3/3): xorg-x11-drv-nvidia-libs-455.38-1.fc31.x86_64.rpm 2.2 MB/s | 78 MB 00:35

Total 3.4 MB/s | 121 MB 00:35
Running transaction check
Transaction check succeeded.
Running transaction test
Transaction test succeeded.
Running transaction
Preparing : 1/1
Installing : xorg-x11-drv-nvidia-libs-3:455.38-1.fc31.x86_64 1/3
Installing : xorg-x11-drv-nvidia-kmodsrc-3:455.38-1.fc31.x86_64 2/3
Installing : xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 3/3
Running scriptlet: xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 3/3
Verifying : xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 1/3
Verifying : xorg-x11-drv-nvidia-kmodsrc-3:455.38-1.fc31.x86_64 2/3
Verifying : xorg-x11-drv-nvidia-libs-3:455.38-1.fc31.x86_64 3/3

Installed:
xorg-x11-drv-nvidia-cuda-libs-3:455.38-1.fc31.x86_64 xorg-x11-drv-nvidia-kmodsrc-3:455.38-1.fc31.x86_64
xorg-x11-drv-nvidia-libs-3:455.38-1.fc31.x86_64

Skipped:
akmod-nvidia-3:455.38-1.fc31.x86_64
Complete!

Bottom line: I have installed

dnf list installed ‘nvidia’
Installed Packages
nvidia-query-resource-opengl.x86_64 1.0.0-6.fc31 @fedora
nvidia-query-resource-opengl-lib.x86_64 1.0.0-6.fc31 @fedora
nvidia-texture-tools.x86_64 2.0.8-21.fc31 @fedora
nvidia-texture-tools-devel.x86_64 2.0.8-21.fc31 @fedora
xorg-x11-drv-nvidia-cuda-libs.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-kmodsrc.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-libs.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver

missing

akmod-nvidia.x86_64                                                          3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
nvidia-persistenced.x86_64                                                   3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
nvidia-settings.x86_64                                                       3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia.x86_64                                                   3:455.38-1.fc32                                     @rpmfusion-nonfree-updates
xorg-x11-drv-nvidia-cuda.x86_64

This may be a change from fed 31 to fed32. Maybe upgrade to 32 and try again or go all the way to 33?
I don’t like the module filter messages I was getting.

computersavvy · November 24, 2020, 1:26am

I am not sure what is different there for you.
I have never used the

repo and have been using nvidia drivers from rpmfusion on fedora since at least F27 with no problems whatever.
In fact my current repolist is (and has been for some time)

fedora                                                                            Fedora 32 - x86_64
fedora-cisco-openh264                                                             Fedora 32 openh264 (From Cisco) - x86_64
google-chrome                                                                     google-chrome
rpmfusion-free                                                                    RPM Fusion for Fedora 32 - Free
rpmfusion-free-updates                                                            RPM Fusion for Fedora 32 - Free - Updates
rpmfusion-nonfree                                                                 RPM Fusion for Fedora 32 - Nonfree
rpmfusion-nonfree-steam                                                           RPM Fusion for Fedora 32 - Nonfree - Steam
rpmfusion-nonfree-updates                                                         RPM Fusion for Fedora 32 - Nonfree - Updates
updates                                                                           Fedora 32 - x86_64 - Updates

The only thing I see causing the problem is the modular issue. Something was installed there that needs to be removed to eliminate the conflict. It is preventing the install of the xorg package and may be related to the nvidia-kmod-common package that is not used by F32. This site gives some instructions about modules in fedora 31.
Try removing nvidia-kmod-common then installing those packages you listed as missing.

If this doesn’t work we will have to try something else.

olinart · November 25, 2020, 3:24am

Yeah, that’s where I’ve been stuck. I can’t find nvidia-kmod-common or anything like it in the dnf listings

search nvidia-kmod
Last metadata expiration check: 3:27:11 ago on Tue 24 Nov 2020 03:34:40 PM PST.
========================== Name Matched: nvidia-kmod ===========================
xorg-x11-drv-nvidia-kmodsrc.x86_64 : xorg-x11-drv-nvidia kernel module source
: code

in downloads outside of dnf

r> pm -qa |grep nvidia

xorg-x11-drv-nvidia-kmodsrc-455.38-1.fc31.x86_64
xorg-x11-drv-nvidia-libs-455.38-1.fc31.x86_64
xorg-x11-drv-nvidia-cuda-libs-455.38-1.fc31.x86_64
nvidia-texture-tools-devel-2.0.8-21.fc31.x86_64
nvidia-texture-tools-2.0.8-21.fc31.x86_64
nvidia-query-resource-opengl-1.0.0-6.fc31.x86_64
nvidia-query-resource-opengl-lib-1.0.0-6.fc31.x86_64

Nor is it an installed module

dnf module list
Last metadata expiration check: 1:54:07 ago on Tue 24 Nov 2020 03:34:40 PM PST.
@modulefailsafe
Name Stream Profiles Summary
gimp 2.10 [e] default, dev GIMP
el
meson latest [e] default The Meson Build system
ninja latest [e] default Small build system with a focus on sp
eed
nvidia-driver latest-dkms [e] default, fm, Nvidia driver for latest-dkms branch
ks

Hint: [d]efault, [e]nabled, disabled, [i]nstalled
So not surprisingly dnf remove nvidia-kmod-common fails, and yet it stops me from loading the driver.
Even dnf autoremove and dnf distro-sync didn’t clear this problem.
Do you think upgrading to 32 or 33 might help? At least it is probably neutral. Or I could download the 455.38 driver from nvidia. The akmod-nvidia is still a problem then.

computersavvy · November 26, 2020, 8:48pm

This from your modules list is likely the problem. try removing that then finish the nvidia install while using only the repos already listed. The error seemed modules related anyway.

olinart · November 27, 2020, 1:00am

I already tried that following the modules link you sent. The (e) status means that it is enabled but not installed, which is apparently correct for a removed module.
I’ve basically scoured my disk for anything relating to the dependency and filtering messages and come up empty.
During the last year on f31 the 450 nvidia driver was the default with no support for cuda 11, evolved to the 455 driver with cuda11 support that I got working this summer. Probably would have been more straightforward if I had upgraded to f32 instead.

computersavvy · November 27, 2020, 2:16am

I think that page I linked suggested disabling what you did not want. Just having it enabled might be the issue.

If that does not work then a clean install of either F32 or 33 should fix it. I would go with 32 first just because I know it has worked for me for more than 6 months. I once had a similar problem with gimp when I left the modular repos enabled and had to explicitly remove the module to stop the errors.

olinart · November 27, 2020, 9:43pm

Thanks, Jeffrey!

dnf module reset nvidia-driver

cleared the module problem. I was able to install the akmod and driver.

dnf list installed ‘nvidia’
Installed Packages
akmod-nvidia.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
kmod-nvidia-5.8.18-100.fc31.x86_64.x86_64 3:455.38-1.fc31 @@commandline
nvidia-persistenced.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
nvidia-query-resource-opengl.x86_64 1.0.0-6.fc31 @fedora
nvidia-query-resource-opengl-lib.x86_64 1.0.0-6.fc31 @fedora
nvidia-settings.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
nvidia-texture-tools.x86_64 2.0.8-21.fc31 @fedora
nvidia-texture-tools-devel.x86_64 2.0.8-21.fc31 @fedora
xorg-x11-drv-nvidia.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-cuda.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-cuda-libs.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-devel.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-kmodsrc.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver
xorg-x11-drv-nvidia-libs.x86_64 3:455.38-1.fc31 @rpmfusion-nonfree-nvidia-driver

and a reboot shows me using the nvidia driver.
I still have some work to do with the xorg to get monitors attached to the gpu card working, but this is a big step forward. Thanks for your expert advice.

computersavvy · November 28, 2020, 2:04am

Glad to be of assistance and good luck with the rest of the issue.

olinart · December 28, 2020, 7:58pm

Just reporting that I was able to upgrade to f32 following your advice on a now-closed thread (curl-error) after dnf erasing the nvidia drivers and then reinstalling them. This worked with only a minor hitch, and I get the 455.45.01-1 drivers. I had to reinstall the /etc/X11/xorg.conf.d/nvidia.conf file with mods for the external gpu:

#This file is provided by xorg-x11-drv-nvidia mod AO
#Do not edit

Section “OutputClass”
Identifier “nvidia”
MatchDriver “nvidia-drm”
Driver “nvidia”
Option “AllowEmptyInitialConfiguration”
Option “AllowExternalGpus”
Option “SLI” “Auto”
Option “BaseMosaic” “on”
EndSection

Section “ServerLayout”
Identifier “layout”
Option “AllowNVIDIAGPUScreens”
EndSection

Your advice to not upgrade directly to f33 was also good. I get the following warnings from a f32-f33 upgrade. Thanks also for that one. I’m concerned about the system-config-users , dnf-yum, lsvpd broken reports that should have nothing to do with gstreamer / rpmfusion stuff. Is it safe to proceed with skip-broken or do I need some other repo enabled?

No match for group package “lsvpd”
No match for group package “powerpc-utils”
No match for group package “xplayer”
No match for group package “nemo-extension-xreader”
No match for group package “exaile”
No match for group package “dnf-yum”
No match for group package “pix”
No match for group package “fedora-release-notes”
No match for group package “xviewer”
No match for group package “kexec-tools-anaconda-addon”
No match for group package “system-config-users”
Error:
Problem: package gstreamer1-plugins-ugly-free-devel-1.16.2-2.fc32.x86_64 requires gstreamer1-plugins-ugly-free = 1.16.2-2.fc32, but none of the providers can be installed

gstreamer1-plugins-ugly-free-1.16.2-2.fc32.x86_64 does not belong to a distupgrade repository

problem with installed package gstreamer1-plugins-ugly-free-devel-1.16.2-2.fc32.x86_64

computersavvy · December 29, 2020, 5:51am

The gstreamer1-plugins-ugly-free and gstreamer1-plugins-ugly-free-devel packages were possibly installed from rpmfusion and they are now provided directly from fedora. simply erase those 2 packages and re-install after the upgrade is done. You may need to allow it to erase additional packages as well, although erasing most dependencies can be avoided with the --noautoremove option. If it removes other packages keep a note of them and re-install them as well.

The messages about the group packages are a result of having the modular repos enabled. The fix should be a simple command to (semi)permanently disable those repos during the upgrade.

 sudo dnf repolist

which will list all the repos you have enabled.
Then to disable the modular repos if they are seen with the command above and you choose to do so.

sudo dnf config-manager --set-disabled fedora-modular 
sudo dnf config-manager --set-disabled updates-modular

This change can be reversed by changing the “–set-disabled” to “–set-enabled” with a repeat of the command.

system · January 26, 2021, 5:51am

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.